---
title: "Can a 1B Model Trained Only on Legal Data Compete? Denmark's Mimir v1 Says Yes"
date: 2026-08-30
originalDate: 2026-08-15
originalTitle: "1B小模型打赢大模型？丹麦人只喂\"干净合法\"的数据"
issue: "Batch 2"
description: "One billion parameters, 161 permissible datasets, no copyright gray zones — and state-of-the-art Danish plus near-parity with 4B models on English."
tags:
  - "small models"
  - "permissible data"
  - "HRM architecture"
  - "open weights"
  - "low-resource languages"
papers:
  - id: "2608.13517"
    title: "DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data"
    url: "https://arxiv.org/abs/2608.13517"
sources:
  - label: "Hugging Face: danish-foundation-models/DFM-Mimir (open weights)"
    url: "https://huggingface.co/danish-foundation-models/DFM-Mimir"
  - label: "Danish Foundation Models: Mimir v1 release note"
    url: "https://www.foundationmodels.dk/news/2026/08/14/mimir-1-release-note.html"
faq:
  - q: "What is Mimir v1?"
    a: "A 1-billion-parameter open model from the Danish Foundation Models project, built on the HRM-Text architecture (the Hierarchical Reasoning Model's dual-loop recurrence rather than stacked Transformers). Its defining constraint: instruction/post-training exclusively on legally permissible data — 161 datasets, no copyright gray zones, no distillation from frontier models. Weights and data are open for download, inspection and reproduction."
  - q: "How well does it actually perform?"
    a: "Across 20 benchmarks (English, math, code, Danish): it beats its same-architecture predecessor HRM-Text 1B, competes with models several times its size (Qwen 3.5 4B, Gemma 4 E2B — on English within ~0.3 points of Qwen 3.5 4B despite being 4x smaller), and sets a new state of the art for Danish among open models. The boundary the paper itself draws: a strong small model, not a leaderboard summit — 1B is still 1B."
  - q: "Why does 'permissible data only' matter?"
    a: "The industry's open secret is that frontier training sets are copyright minefields. Large labs absorb lawsuits; small teams and research institutions cannot. Mimir's existence proof: the shackles do not prevent dancing — model floors are held up by data quality and architecture, not parameter count alone. For small-language communities the point doubles: Danish cannot wait for giants to allocate it attention; done locally with compliant data, it can be best-in-class at home."
  - q: "What is the practical reading for teams?"
    a: "The column's takeaway for developers: don't be intimidated by 'no GPUs, no data, no seat at the table' — clean data plus a suitable architecture plus a defined scenario (local deployment, private data, vertical tasks) is a viable niche. For everyone else: the model market is not inevitably winner-take-all; compliant, open, small-and-beautiful players are finding their ecology. The uncomfortable question the piece leaves: if permissible-only models forever trail anything-goes models by 5%, which does the market pick — and which do you?"
---
The arms-race logic of large models is simple: more data, more parameters, copyright later. On August 13–14, 2026, a Danish preprint demonstrated another way to live: one billion parameters, training data entirely legal, performance nonetheless approaching models several times larger. Our Chinese-language column's piece ran under "A 1B small model beating big ones? The Danes only fed it 'clean and legal' data" — the Nordic-kitchen joke writes itself: clean ingredients, real craft. This digest translates it; the paper, weights and release note are linked and verified.

## The claims

- 1B parameters on HRM-Text — the Hierarchical Reasoning Model lineage, not the standard Transformer stack
- 161 datasets in the permissible-only instruction/post-training mixture; trained from the base with no frontier-model distillation
- 20 benchmarks: beats HRM-Text 1B; competes with Qwen 3.5 4B and Gemma 4 E2B; ~0.3 points behind Qwen 3.5 4B on English at one-quarter the size; state of the art for Danish among open models
- Everything released: weights on Hugging Face, release note at foundationmodels.dk

## How the Chinese column reads it

The framing device is the kitchen: with clean, legal ingredients the dish approaches a master chef's — through craft (architecture and training method), not ingredient volume. Why the route deserves its own story: for teams that cannot absorb copyright litigation, "must use clean data" had meant dancing in shackles; Mimir proves the shackles don't prevent dancing. And for small-language countries the move is existential rather than virtuous — Danish is an edge language in the giants' economics; done domestically with compliant data it becomes best-in-class. The column's sharpest line: **data compliance is not a virtue performance; it is a survival strategy** — when the copyright bill actually comes due, the clean companies sleep at night.

The cold water is administered honestly: Mimir v1 is "strong within small parameters," not "top of the general board" — competing with 4B-class models is not flogging frontier ones; physics remains. The value is the proof, not the summit: model capability floors are propped up by data quality and architecture design, not parameter count alone.

## What to watch

Whether the permissible-data route compounds (more teams, more architectures, better craft) or plateaus; whether regulator or litigation timelines turn "clean" from choice to requirement — which would re-price every model trained the other way; and whether the small-language pattern replicates (one country, one compliant corpus, one open model) as a counter-cartography to the frontier labs.

> **Provenance & disclosure.** Originally published in Chinese on our WeChat channel on 2026-08-15 ("1B小模型打赢大模型？丹麦人只喂'干净合法'的数据"); drafted with AI assistance under human editorial direction. Translated to English on 2026-08-30 (AI-assisted, human-reviewed). Paper ID, benchmark claims and openness verified against the arXiv abstract, the Hugging Face model card and the release note; the column's "trained from scratch" phrasing narrowed to "permissible post-training mixture" per the paper's own title. Preprint caveat inherited. Translated commentary — not a SigPulse measurement; first-party numbers live in the [dispatches](/posts/) and the [/data/ ledger](/data/).